Japanese Smart Home Command Normalizer

Normalize short Japanese smart-home voice transcripts after STT into safe structured intents and slots. Use when handling Japanese commands for lights and ai...

Security Scan

The skill's code, instructions, and metadata are coherent with its stated purpose (Japanese smart-home transcript normalization) and do not request unrelated credentials, system paths, or network installs.

Purpose & Capability

Name/description match included code: a normalization/classification module, sample fixtures, tests, and integration notes. It does not request unrelated env vars or binaries and only claims to output normalized intents for downstream device-control.

Instruction Scope

SKILL.md confines runtime behavior to reading bundled reference files, using the included lib/normalize.js, and optionally running the demo script. It explicitly states it does not call device APIs and instructs returning needsConfirmation for low confidence.

Install Mechanism

No install spec; package contains only local JS modules, tests, fixtures, and docs. No network downloads, external installers, or extract steps are present.

Credentials

The skill requires no environment variables, credentials, or config paths. All processing uses local code and bundled fixtures.

Persistence & Privilege

Flags show default behavior (not always:true). The skill does not request permanent presence or modify other skills; it can be invoked by the agent (normal) but does not itself persist credentials or change system settings.

Assessment

This skill appears to be a local text-normalizer for short Japanese smart-home transcripts and does not access external services or request secrets. Before installing, verify how you will connect it to a device-control skill: the normalizer intentionally does not execute device APIs, so ensure the downstream executor enforces confirmation and authorization. If you plan to run the demo or tests, note they read local fixture files only. If you need stricter autonomy controls, disable automatic invocation or audit the downstream device-control skill that will act on the normalized intents.

Use this skill when a short Japanese STT transcript needs to be normalized before smart-home execution.

Workflow

1. Read references/design.md for the normalization pipeline and result shape.
2. Read references/domains.md for the supported domains and vocabulary.
3. Reuse lib/normalize.js as the core pure module.
4. Use scripts/demo.js to try sample transcripts from the terminal.
5. Integrate the normalized result into a device-control skill such as switchbot-light or a hook such as audio-router.

Current domains

• light
  • device aliases: 電気, ライト, 照明
  • actions: on, off
• aircon
  • device aliases: エアコン
  • actions: on, off, set_mode
  • modes: cool, heat, dry, fan

Notes

• This skill only normalizes and classifies text. It does not call device APIs.
• Prefer fixed vocabulary plus lightweight fuzzy matching over open-ended LLM interpretation for safety-critical home actions.
• When confidence is low or required slots are missing, return needsConfirmation: true instead of auto-executing.
• Add future devices by extending the domain vocabulary, not by piling more ad-hoc regex into callers.

Resources

• lib/normalize.js: core normalization and classification module.
• scripts/demo.js: print normalized results for sample inputs.
• fixtures/samples.json: sample transcripts and expected outcomes.
• references/design.md: pipeline, API shape, and confidence rules.
• references/domains.md: supported vocabulary and extension guidance.
• references/openclaw-integration.md: thin-hook integration guidance.